Optimal F better than Kelly in trading???

Discussion in 'Risk Management' started by chrisbamtrader123, Feb 4, 2021.

  1. Just wondering if anyone had any thoughts on this? I have 14 years trading experience but unfortunately it was long in to my trading experience before I realised the paramount importance of sound money management. Just wondering if any one has successful personal experience of using optimal f. I know Ralph Vince believes it to be far superior to Kelly in trading. Would be grateful to hear any ones viewpoint?
     
  2. nothing special about optimal f vs kelly, optimal f is simply a constraint on kelly normalizing it to fall btwn 0 and 1

    both are silly to use at full values as any inputs we use are subject to tremendous uncertainty, estimation error, and non stationarity. even Ralph vince acknowledges this in his latest book LSM which is essentially him admitting the fragility of geometric sizing and advocating a more robust approach

    optimal f / kelly is best used as an indicator of trade quality or as your upper limit on leverage. terms like 'half kelly' and 'quarter kelly' are often thrown around but are completely arbitrary, simulations are really your best way to size your bets. I think euan Sinclair offers a few additional formulas in his book vol trading where you can assess the optimal kelly fraction as a probability of some level of DD.

    also important to note is your progression scheme, kelly traditionally proposes an anti-martingale scheme where you size up after a win and scale down after a loss. however (contrary to what some keyboard experts have you believe) there are times when a negative progression scheme is more appropriate and you size up after a loss and size down after a win. theres more to it but I dont wanna get too into the weeds, cc Aaron brown and Sid Browne for their research on this
     
    Russell Shuffle, drm7 and guru like this.
  3. Collagen

    Collagen

    What Ducatista said. Vince book is extremely interesting to read although not very easy. Missing point of Optimal F is in my opinion the fact that We don't know what to expect as we run into the future, aka if our systems will hold. (meaning that if the systems/portfolio doesn't hold MaxDD of the past (backtest-OOS etc) we will trade well above optimal F causing a giant margin call. The other point I have is that Optimal F is based on a fixed $ amount while there are probably more appropriate way to handle risk/return.

    The second part Ducatista mentioned which is equally interesting is called dependency. Dependency may exist or not and there a couple of way to test it. It's a bit dangerous meaning that We don't know exactly if a positive/negative dependency in the past will translate in the same outcome into the future.
     
  4. qlai

    qlai

  5. ValeryN

    ValeryN

    Problem with those is quality of input. Garbage in garbage out. Most systems don't have too many trades to show sufficient variety of that can go wrong. I personally look at the nature of trade and size up not according to backtest DD but rather what's the worst case for the underlying.

    For examples - stocks do gap down overnight. If long system trades stocks it is just a matter of time it will stumble upon something that gaps down 50-70%. Regardless of what backtest shows. With shorts - they can 10x over a few days. There are lots of assumptions one can make about risk for particular industry or market cap but that's the general idea I use.

    Val
     
  6. Justrade

    Justrade

    One thing I have learned in trading is that your largest loss and/or drawdown is always ahead of you
     
    Last edited: Feb 4, 2021
    hypercube likes this.
  7. Just echoing what ducatista said. Kelly is problematic in a lot of ways, as is, as he points out, the scaling scheme. Some of that boils down to "how" you determine the side of a trade. Is it trend, momentum, reversion etc. Those will have you taking a different approach to how and when you scale (size of the bet).
     
  8. so fixed ratio by ryan jones is good for compounding the account to grow it much quicker?
     
  9. but what about position sizing? If Kelly is not the best way which I agree I dont think it is for trading....what have you guys found the best way to calculate optimum position size?
     
  10. modeling residual errors to estimate probability of true-positive outcomes.
    That's the last step in a pretty long pipe-line though, you've got to have a lot of other things right, before you get there.
     
    #10     Feb 5, 2021